Increase ping delay in WatchdogActorTest to try and reduce spurious test failures...
[arvados.git] / services / nodemanager / tests / test_computenode_driver_gce.py
1 #!/usr/bin/env python
2
3 from __future__ import absolute_import, print_function
4
5 import json
6 import time
7 import unittest
8
9 import mock
10
11 import arvnodeman.computenode.driver.gce as gce
12 from . import testutil
13
14 class GCEComputeNodeDriverTestCase(testutil.DriverTestMixin, unittest.TestCase):
15     TEST_CLASS = gce.ComputeNodeDriver
16
17     def setUp(self):
18         super(GCEComputeNodeDriverTestCase, self).setUp()
19         self.driver_mock().list_images.return_value = [
20             testutil.cloud_object_mock('testimage', selfLink='image-link')]
21         self.driver_mock().ex_list_disktypes.return_value = [
22             testutil.cloud_object_mock(name, selfLink=name + '-link')
23             for name in ['pd-standard', 'pd-ssd', 'local-ssd']]
24         self.driver_mock.reset_mock()
25
26     def new_driver(self, auth_kwargs={}, list_kwargs={}, create_kwargs={}):
27         create_kwargs.setdefault('image', 'testimage')
28         return super(GCEComputeNodeDriverTestCase, self).new_driver(
29             auth_kwargs, list_kwargs, create_kwargs)
30
31     def test_driver_instantiation(self):
32         kwargs = {'user_id': 'foo'}
33         driver = self.new_driver(auth_kwargs=kwargs)
34         self.assertTrue(self.driver_mock.called)
35         self.assertEqual(kwargs, self.driver_mock.call_args[1])
36
37     def test_create_image_loaded_at_initialization_by_name(self):
38         image_mocks = [testutil.cloud_object_mock(c) for c in 'abc']
39         list_method = self.driver_mock().list_images
40         list_method.return_value = image_mocks
41         driver = self.new_driver(create_kwargs={'image': 'b'})
42         self.assertEqual(1, list_method.call_count)
43
44     def test_create_includes_ping_secret(self):
45         arv_node = testutil.arvados_node_mock(info={'ping_secret': 'ssshh'})
46         driver = self.new_driver()
47         driver.create_node(testutil.MockSize(1), arv_node)
48         metadata = self.driver_mock().create_node.call_args[1]['ex_metadata']
49         self.assertIn('ping_secret=ssshh', metadata.get('arv-ping-url'))
50
51     def test_create_raises_but_actually_succeeded(self):
52         arv_node = testutil.arvados_node_mock(1, hostname=None)
53         driver = self.new_driver()
54         nodelist = [testutil.cloud_node_mock(1)]
55         nodelist[0].name = 'compute-000000000000001-zzzzz'
56         self.driver_mock().list_nodes.return_value = nodelist
57         self.driver_mock().create_node.side_effect = IOError
58         n = driver.create_node(testutil.MockSize(1), arv_node)
59         self.assertEqual('compute-000000000000001-zzzzz', n.name)
60
61     def test_create_sets_default_hostname(self):
62         driver = self.new_driver()
63         driver.create_node(testutil.MockSize(1),
64                            testutil.arvados_node_mock(254, hostname=None))
65         create_kwargs = self.driver_mock().create_node.call_args[1]
66         self.assertEqual('compute-0000000000000fe-zzzzz',
67                          create_kwargs.get('name'))
68         self.assertEqual('dynamic.compute.zzzzz.arvadosapi.com',
69                          create_kwargs.get('ex_metadata', {}).get('hostname'))
70
71     def test_create_tags_from_list_tags(self):
72         driver = self.new_driver(list_kwargs={'tags': 'testA, testB'})
73         driver.create_node(testutil.MockSize(1), testutil.arvados_node_mock())
74         self.assertEqual(['testA', 'testB'],
75                          self.driver_mock().create_node.call_args[1]['ex_tags'])
76
77     def test_create_with_two_disks_attached(self):
78         driver = self.new_driver(create_kwargs={'image': 'testimage'})
79         driver.create_node(testutil.MockSize(1), testutil.arvados_node_mock())
80         create_disks = self.driver_mock().create_node.call_args[1].get(
81             'ex_disks_gce_struct', [])
82         self.assertEqual(2, len(create_disks))
83         self.assertTrue(create_disks[0].get('autoDelete'))
84         self.assertTrue(create_disks[0].get('boot'))
85         self.assertEqual('PERSISTENT', create_disks[0].get('type'))
86         init_params = create_disks[0].get('initializeParams', {})
87         self.assertEqual('pd-standard-link', init_params.get('diskType'))
88         self.assertEqual('image-link', init_params.get('sourceImage'))
89         # Our node images expect the SSD to be named `tmp` to find and mount it.
90         self.assertEqual('tmp', create_disks[1].get('deviceName'))
91         self.assertTrue(create_disks[1].get('autoDelete'))
92         self.assertFalse(create_disks[1].get('boot', 'unset'))
93         self.assertEqual('SCRATCH', create_disks[1].get('type'))
94         init_params = create_disks[1].get('initializeParams', {})
95         self.assertEqual('local-ssd-link', init_params.get('diskType'))
96
97     def test_list_nodes_requires_tags_match(self):
98         # A node matches if our list tags are a subset of the node's tags.
99         # Test behavior with no tags, no match, partial matches, different
100         # order, and strict supersets.
101         cloud_mocks = [
102             testutil.cloud_node_mock(node_num, tags=tag_set)
103             for node_num, tag_set in enumerate(
104                 [[], ['bad'], ['good'], ['great'], ['great', 'ok'],
105                  ['great', 'good'], ['good', 'fantastic', 'great']])]
106         cloud_mocks.append(testutil.cloud_node_mock())
107         self.driver_mock().list_nodes.return_value = cloud_mocks
108         driver = self.new_driver(list_kwargs={'tags': 'good, great'})
109         self.assertItemsEqual(['5', '6'], [n.id for n in driver.list_nodes()])
110
111     def build_gce_metadata(self, metadata_dict):
112         # Convert a plain metadata dictionary to the GCE data structure.
113         return {
114             'kind': 'compute#metadata',
115             'fingerprint': 'testprint',
116             'items': [{'key': key, 'value': metadata_dict[key]}
117                       for key in metadata_dict],
118             }
119
120     def check_sync_node_updates_hostname_tag(self, plain_metadata):
121         start_metadata = self.build_gce_metadata(plain_metadata)
122         arv_node = testutil.arvados_node_mock(1)
123         cloud_node = testutil.cloud_node_mock(
124             2, metadata=start_metadata.copy(),
125             zone=testutil.cloud_object_mock('testzone'))
126         self.driver_mock().ex_get_node.return_value = cloud_node
127         driver = self.new_driver()
128         driver.sync_node(cloud_node, arv_node)
129         args, kwargs = self.driver_mock().ex_set_node_metadata.call_args
130         self.assertEqual(cloud_node, args[0])
131         plain_metadata['hostname'] = 'compute1.zzzzz.arvadosapi.com'
132         self.assertEqual(
133             plain_metadata,
134             {item['key']: item['value'] for item in args[1]})
135
136     def test_sync_node_updates_hostname_tag(self):
137         self.check_sync_node_updates_hostname_tag(
138             {'testkey': 'testvalue', 'hostname': 'startvalue'})
139
140     def test_sync_node_adds_hostname_tag(self):
141         self.check_sync_node_updates_hostname_tag({'testkey': 'testval'})
142
143     def test_sync_node_raises_exception_on_failure(self):
144         arv_node = testutil.arvados_node_mock(8)
145         cloud_node = testutil.cloud_node_mock(
146             9, metadata={}, zone=testutil.cloud_object_mock('failzone'))
147         mock_response = self.driver_mock().ex_set_node_metadata.side_effect = (Exception('sync error test'),)
148         driver = self.new_driver()
149         with self.assertRaises(Exception) as err_check:
150             driver.sync_node(cloud_node, arv_node)
151         self.assertIs(err_check.exception.__class__, Exception)
152         self.assertIn('sync error test', str(err_check.exception))
153
154     def test_node_create_time_zero_for_unknown_nodes(self):
155         node = testutil.cloud_node_mock()
156         self.assertEqual(0, gce.ComputeNodeDriver.node_start_time(node))
157
158     def test_node_create_time_for_known_node(self):
159         node = testutil.cloud_node_mock(metadata=self.build_gce_metadata(
160                 {'booted_at': '1970-01-01T00:01:05Z'}))
161         self.assertEqual(65, gce.ComputeNodeDriver.node_start_time(node))
162
163     def test_node_create_time_recorded_when_node_boots(self):
164         start_time = time.strftime('%Y-%m-%dT%H:%M:%SZ', time.gmtime())
165         arv_node = testutil.arvados_node_mock()
166         driver = self.new_driver()
167         driver.create_node(testutil.MockSize(1), arv_node)
168         metadata = self.driver_mock().create_node.call_args[1]['ex_metadata']
169         self.assertLessEqual(start_time, metadata.get('booted_at'))
170
171     def test_known_node_fqdn(self):
172         name = 'fqdntest.zzzzz.arvadosapi.com'
173         node = testutil.cloud_node_mock(metadata=self.build_gce_metadata(
174                 {'hostname': name}))
175         self.assertEqual(name, gce.ComputeNodeDriver.node_fqdn(node))
176
177     def test_unknown_node_fqdn(self):
178         # Return an empty string.  This lets fqdn be safely compared
179         # against an expected value, and ComputeNodeMonitorActor
180         # should try to update it.
181         node = testutil.cloud_node_mock(metadata=self.build_gce_metadata({}))
182         self.assertEqual('', gce.ComputeNodeDriver.node_fqdn(node))
183
184     def test_deliver_ssh_key_in_metadata(self):
185         test_ssh_key = 'ssh-rsa-foo'
186         arv_node = testutil.arvados_node_mock(1)
187         with mock.patch('__builtin__.open',
188                         mock.mock_open(read_data=test_ssh_key)) as mock_file:
189             driver = self.new_driver(create_kwargs={'ssh_key': 'ssh-key-file'})
190         mock_file.assert_called_once_with('ssh-key-file')
191         driver.create_node(testutil.MockSize(1), arv_node)
192         metadata = self.driver_mock().create_node.call_args[1]['ex_metadata']
193         self.assertEqual('root:ssh-rsa-foo', metadata.get('sshKeys'))
194
195     def test_create_driver_with_service_accounts(self):
196         service_accounts = {'email': 'foo@bar', 'scopes': ['storage-full']}
197         srv_acct_config = {'service_accounts': json.dumps(service_accounts)}
198         arv_node = testutil.arvados_node_mock(1)
199         driver = self.new_driver(create_kwargs=srv_acct_config)
200         driver.create_node(testutil.MockSize(1), arv_node)
201         self.assertEqual(
202             service_accounts,
203             self.driver_mock().create_node.call_args[1]['ex_service_accounts'])
204
205     def test_fix_string_size(self):
206         # As of 0.18, the libcloud GCE driver sets node.size to the size's name.
207         # It's supposed to be the actual size object.  Make sure our driver
208         # patches that up in listings.
209         size = testutil.MockSize(2)
210         node = testutil.cloud_node_mock(size=size)
211         node.size = size.name
212         self.driver_mock().list_sizes.return_value = [size]
213         self.driver_mock().list_nodes.return_value = [node]
214         driver = self.new_driver()
215         nodelist = driver.list_nodes()
216         self.assertEqual(1, len(nodelist))
217         self.assertIs(node, nodelist[0])
218         self.assertIs(size, nodelist[0].size)
219
220     def test_skip_fix_when_size_not_string(self):
221         # Ensure we don't monkeypatch node sizes unless we need to.
222         size = testutil.MockSize(3)
223         node = testutil.cloud_node_mock(size=size)
224         self.driver_mock().list_nodes.return_value = [node]
225         driver = self.new_driver()
226         nodelist = driver.list_nodes()
227         self.assertEqual(1, len(nodelist))
228         self.assertIs(node, nodelist[0])
229         self.assertIs(size, nodelist[0].size)
230
231     def test_node_found_after_timeout_has_fixed_size(self):
232         size = testutil.MockSize(4)
233         cloud_node = testutil.cloud_node_mock(size=size.id)
234         self.check_node_found_after_timeout_has_fixed_size(size, cloud_node)
235
236     def test_list_empty_nodes(self):
237         self.driver_mock().list_nodes.return_value = []
238         self.assertEqual([], self.new_driver().list_nodes())